The selection of one of two concurrent speech messages for comprehension was investigated in healthy young adults in two event-related potential experiments. The stories were presented from virtual locations located 30 degrees to the left and right azimuth by convolving the speech message by the appropriate head-related transfer function determined for each individual participant. In addition, task irrelevant probe stimuli were presented in rapid sequence from the same virtual locations. In experiment 1, phoneme probes (/da/ voiced by the same talkers as attended and unattended messages) and band-pass filtered noise probes were presented. Phoneme probes coinciding with the attended message gave rise to a fronto-central negativity similar to the Nd-attention effect relative to the phoneme probes coinciding with the unattended speech message, whereas noise probes from the attended message's location showed a more positive frontal ERP response compared to probes from the unattended location resembling the so-called rejection positivity. In experiment 2, phoneme probes (as in exp. 1) and frequency-shifted (+400 Hz) were compared. The latter were characterized by a succession of negative and positive components that were modulated by location. The results suggest that at least two different neural mechanisms contribute to stream segregation in a cocktail-party setting: enhanced neural processing of stimuli matching the attended message closely (indexed by the Nd-effect) and rejection of stimuli that do not match the attended message at the attended location only (indexed by the rejection positivity).